CN117372854A - Real-time detection method for hidden danger diseases of deep water structure of dam - Google Patents
Real-time detection method for hidden danger diseases of deep water structure of dam Download PDFInfo
- Publication number
- CN117372854A CN117372854A CN202311384543.3A CN202311384543A CN117372854A CN 117372854 A CN117372854 A CN 117372854A CN 202311384543 A CN202311384543 A CN 202311384543A CN 117372854 A CN117372854 A CN 117372854A
- Authority
- CN
- China
- Prior art keywords
- dam
- disease
- deep water
- target detection
- water structure
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 201000010099 disease Diseases 0.000 title claims abstract description 73
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 title claims abstract description 73
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 43
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000011897 real-time detection Methods 0.000 title claims abstract description 13
- 238000001514 detection method Methods 0.000 claims abstract description 57
- 238000013138 pruning Methods 0.000 claims abstract description 18
- 238000013140 knowledge distillation Methods 0.000 claims abstract description 11
- 238000012545 processing Methods 0.000 claims abstract description 10
- 238000004590 computer program Methods 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 17
- 238000012549 training Methods 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 7
- 238000003860 storage Methods 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000004821 distillation Methods 0.000 claims description 5
- 210000000988 bone and bone Anatomy 0.000 claims description 4
- 239000000463 material Substances 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 2
- 230000007547 defect Effects 0.000 abstract description 19
- 238000009826 distribution Methods 0.000 abstract description 5
- 230000006835 compression Effects 0.000 abstract description 4
- 238000007906 compression Methods 0.000 abstract description 4
- 238000013508 migration Methods 0.000 abstract description 3
- 230000005012 migration Effects 0.000 abstract description 3
- 230000015556 catabolic process Effects 0.000 abstract description 2
- 238000006731 degradation reaction Methods 0.000 abstract description 2
- 230000003287 optical effect Effects 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000007689 inspection Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000006424 Flood reaction Methods 0.000 description 1
- 229910000831 Steel Inorganic materials 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000003973 irrigation Methods 0.000 description 1
- 230000002262 irrigation Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000007430 reference method Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 230000007847 structural defect Effects 0.000 description 1
- 238000009966 trimming Methods 0.000 description 1
- 208000016261 weight loss Diseases 0.000 description 1
- 230000004580 weight loss Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/05—Underwater scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/096—Transfer learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a real-time detection method for hidden danger diseases of a deep water structure of a dam, which uses a single-stage target detection model YOLOv5s based on an anchor frame as a basic model, constructs a multi-class defect identifier of the dam, changes the weight distribution of a model batch processing layer by utilizing model sparsification, and removes model redundancy parameters according to an optimal pruning strategy. And (3) comprehensively utilizing the model migration and knowledge distillation theory, recovering the problem of precision degradation caused by pruning compression parameters, and finally constructing a light and efficient method for detecting the defects of the underwater structure of the dam. The method can realize real-time target detection of the dam deep water multi-category structure defect diseases under different complex scenes by only taking optical image data as input.
Description
Technical Field
The invention relates to a real-time detection method for hidden danger diseases of a deep water structure of a dam, and belongs to the technical field of non-contact defect detection of the deep water structure of the dam.
Background
At present, 9.8 trillion seats of a reservoir dam have been built in China, and the total reservoir capacity is 8983 hundred million m 3 The installed capacity of water is 3.91 million kilowatts, and the scale and quantity are in the forefront of the world. The projects play an important role in regulating and accumulating rivers, resisting floods, hydroelectric power generation, shipping irrigation and the like, and are an important basic guarantee for national water safety and economic development. However, the dam bodies have the common problems of low construction standard, poor construction process, three-side engineering and the like, and the dam bodies have the common hidden trouble of diseases due to long-term maintenance failure and insufficient post-maintenance management.
The traditional dam structure defect identification method is mainly based on manual inspection, and is supplemented with means such as steel ruler, hammering, test paper and the like. Practice shows that the method has the defects of low detection efficiency, high risk, high missed judgment rate and the like, and the identification result is easy to be influenced by subjective judgment of engineers. In addition, manual inspection is difficult to implement in a deep water structural area of a dam, and the improved frogman underwater detection technology also has the problems of limited submergence depth (generally not exceeding 60 m), narrow submergence operation range, short operation time, high potential risk and the like.
The underwater robot is unmanned carrying equipment which realizes power and signal transmission through an umbilical cable, is operated by a shoreside person and can execute specific underwater operation tasks, and the high-definition visible light camera carried by the underwater robot in recent years is applied to high-resolution image video information acquisition of deep water hidden defects of a dam, so that the limitation of low spatial resolution of the traditional sensor and engineering geophysical prospecting sensing means can be effectively overcome.
However, the deep water part of the dam has complex and changeable structural forms and huge underwater volume, and massive image and video data can be generated by one large-scale deep water detection task of the dam. The method is completely dependent on manual reading and recording means to extract information closely related to structural damage from the mass data, and has the advantages of long time consumption, high cost, high recognition difficulty, high misjudgment rate and high omission rate.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: the real-time detection method for hidden danger diseases of the deep water structure of the dam is provided, a model for detecting multi-class defects of the dam is built by using a YOLOv5s single-stage neural network, the model is thinned and pruned, and a real-time target frame for the multi-class defects of the deep water of the dam under strong background interference is built by using model migration and knowledge distillation theory.
The invention adopts the following technical scheme for solving the technical problems:
a real-time detection method for hidden danger diseases of a deep water structure of a dam comprises the following steps:
step 1, obtaining a disease image video of a deep water structure part of a dam, extracting a disease image from the disease image video, and constructing a multi-category disease data set of the deep water structure of the dam;
step 2, marking the information of disease categories and disease areas on each disease image, and dividing the marked data set into a training set, a verification set and a test set;
step 3, constructing a dam deepwater multi-category disease target detection model by taking a single-stage target detection model Yolov5s as an identifier, pre-training the dam deepwater multi-category disease target detection model by utilizing a training set and a verification set, and evaluating by utilizing a test set to obtain the pre-trained dam deepwater multi-category disease target detection model;
step 4, sparse regularization is carried out on the batch processing layer in the dam deepwater multi-category disease target detection model constructed in the step 3, pruning is carried out on the sparse regularized model according to a pruning rate of 0.5, and a pruned target detection model is obtained;
step 5, taking the dam deepwater multi-category disease target detection model constructed in the step 3 as a teacher network, taking the pruned target detection model obtained in the step 4 as a student network, and carrying out parameter optimization on the pruned target detection model by using a knowledge distillation method to obtain a final target detection model;
and 6, acquiring a real-time disease image of the deep water structure part of the dam, and detecting the real-time disease image by using the final target detection model obtained in the step 5 to obtain the category and position result of the disease.
In the step 1, a visible light camera is carried on the underwater robot to obtain a disease image video of the deep water structure part of the dam, the sampling frequency of the visible light camera is 25 frames per second, and one frame is extracted every five frames and put into a data set, so that a multi-category disease data set of the deep water structure of the dam is obtained.
In the step 2, marking each image in the data set by adopting graphic interface visual marking software Labelme, determining the position of a disease area in the image by utilizing a rectangular frame, wherein marking information comprises coordinates of an upper left corner and a lower right corner of the rectangular frame, obtaining coordinates of a central point of the rectangular frame and the length and the width of the rectangular frame according to the coordinates of the upper left corner and the lower right corner, normalizing the coordinates of the central point of the rectangular frame and the length and the width of the rectangular frame, and storing the normalized coordinates in a json file, and simultaneously storing the category of the disease;
the diseases belong to the categories including cracks, depressions, holes and exposed bone materials.
As a preferred embodiment of the present invention, in the step 4, a calculation formula of sparse regularization is as follows:
wherein D represents a dictionary matrix, μ i Represents the sparse vector, m represents the total number of all samples in the multi-category disease dataset, x i Representing the ith sample in the multi-category disease dataset, lambda represents the sparsification ratio of the sample, I 1 、|||| 2 Respectively 1 norm and 2 norm.
As a preferred embodiment of the present invention, in the step 5, the distillation loss function in the knowledge distillation method is as follows:
L cls ＝μL hard (P s ,y)+(1-μ)L soft (P s ,P t )
wherein L is cls Represents the distillation loss function, μ represents the balance parameter between hard loss and soft loss, L hard (P s Y) represents a hard sample loss function determined from the teacher network output and the tag value, L soft (P s ,P t ) Representing soft sample loss function predicted by teacher network, y represents real label of identification object, P s 、P t Respectively representing the predicted values of the student network and the teacher network.
A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, the processor implementing the steps of the dam deepwater construction hazard real-time detection method as described above when executing the computer program.
A computer readable storage medium storing a computer program which when executed by a processor implements the steps of a dam deepwater construction hazard real-time detection method as described above.
Compared with the prior art, the technical scheme provided by the invention has the following technical effects:
1. according to the method, a YOLOv5s single-stage neural network is used as a basic model, a multi-class defect detection model of a dam is constructed, model sparsification and pruning strategies are utilized, model batch processing layer weight distribution is changed, and model redundancy parameters are removed; further, the model migration and knowledge distillation theory is comprehensively utilized, the problem of precision degradation caused by pruning compression parameters is recovered, a real-time target frame of the multi-class defects of the deep water of the dam under strong background interference is constructed, batch rapid detection of massive deep water detection images and video data of the dam is realized, and the cost and potential risk of manual detection are greatly reduced.
2. The method realizes real-time detection of the deep water defect of the dam and rapidly ascertains the damaged area of the deep water structure of the dam; meanwhile, the manual detection frequency is reduced, the detection cost of the emptying warehouse water is avoided, and the cost is greatly saved.
3. The method can realize batch processing of massive dam deepwater detection video image data by only one-time deep learning model construction, has the remarkable advantages of high monitoring frequency, high calculation processing efficiency, high precision, low misjudgment rate, strong practicability and the like, can be further popularized and applied to the detection of the defects of underwater structures of various wading hydraulic buildings, greatly reduces the manual inspection cost and potential risks, and has wide practical application value.
Drawings
FIG. 1 is a flow chart of a real-time detection method for hidden danger diseases of a deep water structure of a dam;
FIG. 2 is a schematic diagram of a YOLOv5s single-stage object detection network architecture;
FIG. 3 is a diagram of a model pruning and redundant parameter removal process;
FIG. 4 is a graph of model parameters and performance as a function of pruning proportion;
FIG. 5 is a model knowledge distillation and parameter tuning process;
FIG. 6 is a graph comparing the effect of the method of the present invention with the effect of the prior art on detecting the damage to the deep water structure of the dam.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below by referring to the drawings are exemplary only for explaining the present invention and are not to be construed as limiting the present invention.
As shown in FIG. 1, the invention provides a real-time detection method for hidden danger diseases of a deep water structure of a dam, which comprises the following specific steps:
step 1, selecting an underwater structure part of a dam, and carrying out deepwater disease detection by adopting a mode of carrying a visible light camera on an underwater robot. Because of the severe underwater imaging environment, the underwater detection platform usually adopts a whole-course video recording mode for detection, and the original sampling data format is usually video. The sampling frame rate of the camera is usually 25 frames per second, a mode of intercepting and storing one frame every five frames is adopted, video streams are converted into image data, a dam deep water multi-category structural defect data set is built, and four typical defects including cracks, holes, pits and exposed bone materials are covered.
And 2, marking the disease type and the disease area information of the disease image by means of a graphic interface visual marking software Labelme so as to distinguish damaged cells from background cells. The rectangular frame marks are used for determining the positions and the corresponding types of the defects in the image, and the marking information comprises coordinates of an upper left corner and a lower right corner of the detection frame, so that coordinates of a central point of the rectangular detection frame and the length and the width of the detection frame can be deduced accordingly. And further, after normalizing the data information, the class information of the defect belongs to is stored in the JSON file. Disease type numbers are developed from 0, and the types of the underwater diseases of the multi-category dams such as cracks, pits, holes, exposed bone materials and the like are numbered according to corresponding numbers. The acquired image data is divided into a training set, a verification set and a test set according to a certain proportion.
In order to facilitate the construction of a subsequent target detection model, the information release is required to be implemented on the labeled JSON file, and the labeled JSON file is converted into a readable training format, namely an original image file jpg, a detection frame information file xml and disease type label information txt.
And 3, constructing a dam deepwater multi-category defect target detection model by taking a single-stage target detection model YOLOv5s as an identifier, wherein the network architecture of the YOLOv5s is shown in fig. 2. Training a model on a training set, verifying the effectiveness of the model on a verification set, and finally evaluating the generalization capability of the model on a test set to construct a multi-category defect target detection model of the dam deepwater.
And 4, model pruning and redundant parameter removal, as shown in fig. 3. Channel sparsity regularization is an important early work for network weight loss development. The basic purpose is to assign a scale factor to each channel to characterize the channel importance and to use channel sparsification training to distinguish between important and unimportant channels. The model sparsification, pruning and parameter compression means are comprehensively utilized, the weight distribution of the model normalization layer is changed to enable the weight distribution to be mostly close to 0, the pruning means are utilized to remove the redundant parameters and the layer number of the model, and the model reasoning efficiency is improved. The influence of different pruning ratios on the model defect detection performance is tested, and finally, the pruning rate of 0.5 is an important balance point comprehensively considering the parameter scale and the reasoning efficiency of the model, and the model redundancy parameter removal is carried out according to the important balance point, as shown in fig. 4.
The batch layer is the main object of channel sparsification, which is usually located after the convolution layer, and is used to guarantee the fast convergence and generalization capability of the model. The mathematical expression for the batch layer is as follows:
wherein z is in And z out Respectively representing model input and output of the batch layer, e represents noise item of the batch layer, mu B Sum sigma B Respectively representing the mean value and standard deviation of input features in the small batch statistics; gamma and beta represent trainable scale factors and bias parameters, respectively. When the value of the weight coefficient gamma becomes smaller, the corresponding activation function value also decreases. In the convolution-batch module layer, the batch layer is mainly used for realizing channel scaling. By defining scaling factors for each channel in the convolutional layer, sparse regularization can be performed on these scaling factors to automatically determine those channels that are not important, the basic calculation process is as follows:
wherein x is i An i-th sample representing the target dataset; d represents a dictionary matrix; mu (mu) i Representing sparse vector representation; lambda represents the sparsification ratio, which directly determines the regularized sparsification performance of the model, and is an important parameter affecting the sparsification result.
Step 5, knowledge distillation and fine tuning of model parameters, as shown in fig. 5. The combined application of model sparsification and parameter pruning can effectively remove redundant parameters and architecture of a model, but the detection accuracy of a model network is inevitably reduced. To overcome this problem, model trimming and knowledge distillation are combined on the basis of model pruning and redundant parameter removal to eliminate negative effects caused by model pruning and compression.
The fundamental purpose of knowledge distillation is to enhance a lightweight network model (student network) by using supervision information proposed by a high-performance model (teacher network) so as to enable the model to have better detection performance. The teacher network outputs the supervision information as knowledge, and the student network learns to deliver the supervision information is called distillation. The specific calculation is as follows:
L cls ＝μL hard (P s ,y)+(1-μ)L soft (P s ,P t )
wherein L is hard Representing a hard sample loss function judged according to the output of the target detection model and the label value; l (L) soft Representing a soft sample loss function predicted by a teacher network; μ represents the balance parameter between hard and soft losses. The teacher network can better learn the data distribution rule of the data set and obtain better performance in the test set. The soft loss function contains the information about the relation of multiple categories learned by the teacher network, and the learned information can be inserted into the student network through the soft labels.
And 6, arranging a guide line on the structure to be detected, carrying out underwater detection operation according to a planned path, obtaining a real-time disease image of the deep water structure part of the dam, and detecting the real-time disease image by using the final target detection model obtained in the step 5 to obtain the category and position result of the disease.
As shown in fig. 6, under different complex dam deepwater detection scenes such as uneven illumination, obstacle shielding, low visibility and the like, various advanced deep learning target detection algorithms are taken as a reference method, and the method has stronger disease detection generalization capability and robustness under the complex deepwater strong background interference.
Based on the same inventive concept, the embodiment of the application provides a computer device, which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor realizes the steps of the method for detecting hidden danger diseases of the deep water structure of the dam in real time when executing the computer program.
Based on the same inventive concept, the embodiments of the present application provide a computer readable storage medium storing a computer program, which when executed by a processor, implements the steps of the aforementioned method for detecting a hidden danger disease of a deep water structure of a dam in real time.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above embodiments are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereto, and any modification made on the basis of the technical scheme according to the technical idea of the present invention falls within the protection scope of the present invention.
Claims (7)
1. A real-time detection method for hidden danger diseases of a deep water structure of a dam is characterized by comprising the following steps:
step 1, obtaining a disease image video of a deep water structure part of a dam, extracting a disease image from the disease image video, and constructing a multi-category disease data set of the deep water structure of the dam;
step 2, marking the information of disease categories and disease areas on each disease image, and dividing the marked data set into a training set, a verification set and a test set;
step 3, constructing a dam deepwater multi-category disease target detection model by taking a single-stage target detection model Yolov5s as an identifier, pre-training the dam deepwater multi-category disease target detection model by utilizing a training set and a verification set, and evaluating by utilizing a test set to obtain the pre-trained dam deepwater multi-category disease target detection model;
step 4, sparse regularization is carried out on the batch processing layer in the dam deepwater multi-category disease target detection model constructed in the step 3, pruning is carried out on the sparse regularized model according to a pruning rate of 0.5, and a pruned target detection model is obtained;
step 5, taking the dam deepwater multi-category disease target detection model constructed in the step 3 as a teacher network, taking the pruned target detection model obtained in the step 4 as a student network, and carrying out parameter optimization on the pruned target detection model by using a knowledge distillation method to obtain a final target detection model;
and 6, acquiring a real-time disease image of the deep water structure part of the dam, and detecting the real-time disease image by using the final target detection model obtained in the step 5 to obtain the category and position result of the disease.
2. The method for detecting hidden danger diseases of a deep water structure of a dam in real time according to claim 1, wherein in the step 1, an underwater robot is used for carrying a visible light camera to acquire a disease image video of a deep water structure part of the dam, the sampling frequency of the visible light camera is 25 frames per second, and one frame is extracted every five frames and put into a data set, so that a multi-category disease data set of the deep water structure of the dam is obtained.
3. The method for detecting hidden danger diseases of a deep water structure of a dam in real time according to claim 1, wherein in the step 2, each image in a data set is marked by adopting graphic interface visual marking software Labelme, the position of a disease area in the image is determined by utilizing a rectangular frame, marking information comprises coordinates of an upper left corner point and a lower right corner point of the rectangular frame, coordinates of a central point of the rectangular frame and the length and the width of the rectangular frame are obtained according to the coordinates of the upper left corner point and the lower right corner point, the coordinates of the central point of the rectangular frame and the length and the width of the rectangular frame are normalized and then stored in a json file, and meanwhile, the category to which the disease belongs is stored together;
the diseases belong to the categories including cracks, depressions, holes and exposed bone materials.
4. The method for detecting hidden danger diseases of a deep water structure of a dam in real time according to claim 1, wherein in the step 4, a calculation formula of sparse regularization is as follows:
wherein D represents a dictionary matrix, μ i Represents sparse vectors, m representsTotal number of all samples in the multi-category disease dataset, x i Representing the ith sample in the multi-category disease dataset, lambda represents the sparsification ratio of the sample, I 1 、|||| 2 Respectively 1 norm and 2 norm.
5. The method for detecting hidden danger diseases of a deep water structure of a dam in real time according to claim 1, wherein in the step 5, a distillation loss function in the knowledge distillation method is as follows:
L cls ＝μL hard (P s ,y)+(1-μ)L soft (P s ,P t )
wherein L is cls Represents the distillation loss function, μ represents the balance parameter between hard loss and soft loss, L hard (P s Y) represents a hard sample loss function determined from the teacher network output and the tag value, L soft (P s ,P t ) Representing soft sample loss function predicted by teacher network, y represents real label of identification object, P s 、P t Respectively representing the predicted values of the student network and the teacher network.
6. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor, when executing the computer program, performs the steps of the method for real-time detection of a dam deepwater construction risk according to any one of claims 1 to 5.
7. A computer-readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the method for detecting a hidden danger disease of a deep water structure of a dam according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311384543.3A CN117372854A (en) | 2023-10-24 | 2023-10-24 | Real-time detection method for hidden danger diseases of deep water structure of dam |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311384543.3A CN117372854A (en) | 2023-10-24 | 2023-10-24 | Real-time detection method for hidden danger diseases of deep water structure of dam |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117372854A true CN117372854A (en) | 2024-01-09 |
Family
ID=89394279
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311384543.3A Pending CN117372854A (en) | 2023-10-24 | 2023-10-24 | Real-time detection method for hidden danger diseases of deep water structure of dam |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117372854A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117690093A (en) * | 2024-01-31 | 2024-03-12 | 华能澜沧江水电股份有限公司 | Dam safety monitoring operation maintenance method and device, electronic equipment and storage medium |
-
2023
- 2023-10-24 CN CN202311384543.3A patent/CN117372854A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117690093A (en) * | 2024-01-31 | 2024-03-12 | 华能澜沧江水电股份有限公司 | Dam safety monitoring operation maintenance method and device, electronic equipment and storage medium |
| CN117690093B (en) * | 2024-01-31 | 2024-04-26 | 华能澜沧江水电股份有限公司 | Dam safety monitoring operation maintenance method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11521379B1 (en) | Method for flood disaster monitoring and disaster analysis based on vision transformer | |
| CN112308019B (en) | SAR ship target detection method based on network pruning and knowledge distillation | |
| CN109145830B (en) | Intelligent water gauge identification method | |
| CN113705478B (en) | Mangrove single wood target detection method based on improved YOLOv5 | |
| CN111461190A (en) | Deep convolutional neural network-based non-equilibrium ship classification method | |
| CN109740676B (en) | Object detection and migration method based on similar targets | |
| CN109376576A (en) | The object detection method for training network from zero based on the intensive connection of alternately update | |
| CN113408423A (en) | Aquatic product target real-time detection method suitable for TX2 embedded platform | |
| CN113313166B (en) | Ship target automatic labeling method based on feature consistency learning | |
| CN117372854A (en) | Real-time detection method for hidden danger diseases of deep water structure of dam | |
| CN115546565A (en) | YOLOCBF-based power plant key area pipeline oil leakage detection method | |
| CN110599459A (en) | Underground pipe network risk assessment cloud system based on deep learning | |
| CN114511710A (en) | Image target detection method based on convolutional neural network | |
| CN116630285A (en) | Photovoltaic cell type incremental defect detection method based on significance characteristic hierarchical distillation | |
| CN115690568A (en) | Unmanned ship target detection method based on incremental learning | |
| CN112149612A (en) | Marine organism recognition system and recognition method based on deep neural network | |
| CN116486240A (en) | Application of image recognition algorithm in intelligent inspection method of unmanned aerial vehicle of power transmission line | |
| CN111814696A (en) | Video ship target detection method based on improved YOLOv3 | |
| CN116385911A (en) | Lightweight target detection method for unmanned aerial vehicle inspection insulator | |
| CN117789037A (en) | Crop growth period prediction method and device | |
| CN113052103A (en) | Electrical equipment defect detection method and device based on neural network | |
| CN116664846A (en) | Method and system for realizing 3D printing bridge deck construction quality monitoring based on semantic segmentation | |
| CN116682045A (en) | Beam pumping unit fault detection method based on intelligent video analysis | |
| CN116884192A (en) | Power production operation risk early warning method, system and equipment | |
| CN115239034B (en) | Method and system for predicting early defects of wind driven generator blade |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |